Platform Overview
Platform Architecture
How the platform components fit together — what matters for science configuration.
What You Need to Know
| Component | What It Does | What You Control |
|---|---|---|
| AI Interviewer | Retell AI + GPT-4.1 conducts real-time interviews | Interviewer rubric (tone, probing, rules, question generation) |
| AI Scorer | Claude analyzes transcripts post-interview | Scorer rubric (methodology, thresholds, anchors) |
| Competency Framework | Database of competencies with 5-level anchors | Anchor definitions, categories, platform competencies |
| Integrity Monitor | Browser-based face detection + behavior tracking | Deduction weights, detection thresholds |
| AI Helpers | Claude generates JDs, competencies, questions | System/task prompts for generation quality |
Three User Surfaces
- Admin Portal (
/admin) — Where you configure rubrics, competencies, and all science settings - Employer Portal (
/dashboard) — What employers see. Postings, candidates, scorecards, team. - Candidate Portal — Token-gated interview experience. Privacy → resume → interview → complete.
Key Design Decisions
- No audio/video stored — ever. Camera is for live face detection only.
- Scorer never sees candidate identity — name scrubbed from transcript before scoring.
- All science configuration in database —
rubricstable stores JSON configs, editable via admin UI. Nothing is hardcoded. - Soft-delete everywhere — EEOC, NYC 144, GDPR compliance.
- Rubric versioning — every save creates an immutable version snapshot. Versions are stamped on transcripts and scorecards for traceability.